Reproduce “CEDAR: Agent-Orchestrated Tree Search for Goal-Directed Optimization of Complex Systems” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yingtao Tian
CEDAR integrates Monte Carlo Tree Search with Large Language Model agents to autonomously discover and optimize complex systems represented in Python based on natural language goals. This method successfully optimized a World Dynamics system to achieve better trade-offs across population, resources, and pollution metrics, and out-performed black-box optimization in fitting system records.
Key insights
- Integrating LLM agents as intelligent search operators within a principled MCTS framework enables autonomous structural discovery and optimization of complex systems from high-level natural language objectives.
- A unified Python representation, augmented with domain-specific primitives, allows LLMs to effectively manipulate and understand complex system dynamics, reducing the need for specialized modeling languages.
- MCTS with LLMs preserves solution diversity and yields better performance and more stable trajectories compared to linear search or black-box optimization for complex system design.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.